Add a mixed point/wx initialization method

[nwagenbrenner]

This is has been requested by a couple of groups now. Might be more widely used once past casts are enabled.

[jforthofer]

Just a quick note to remind us of how we'd do it when we get to this... You would compute the difference between the measurement and gridded model, then interpolate the differences across the mesh to do a kind of bias correction. Then run the mass conservation solver, and iterate over this process just like in point initialization until the converged field matches at all the points to some tolerance.

[ksshannon]

Or if #247 is implemented, then maybe a freebie?

[jforthofer]

This might be possible with #247 , but I think we'd need to come up with a "feathering" type approach, as described in #247 . What I'm thinking is that blasting the diagonal would force the cell at the measurement point to match, but the cells around it would be relying on the mass conserving solver to "feather" out this to the neighbors. This might not work too well. We may need to come up with a "feathering" approach that decays the effect of the measurement matching out with distance.